Background: The CRASH-3 trial hypothesized that Tranexamic Acid (TXA) could reduce intracranial bleeding and the risk of head injury death in patients with traumatic brain injury (TBI). Because “head injury death” includes death from intracranial bleeding, to simplify the trial procedures, the investigators did not collect data on the extent of intracranial bleeding in all trial patients. Furthermore, TXA may increase the occurrence of stroke, and this outcome was recorded in the trial outcome form, but cerebral infarction as seen on imaging was not. Additional information on the hypothesized mechanism of action of TXA in TBI could help explain the CRASH-3 trial results. Research questions, aims and hypotheses: The CRASH-3 Intracranial Bleeding Mechanistic Study (IBMS) sought to investigate whether the mechanism of action of TXA in TBI could be assessed using routinely collected brain imaging. If so, the IBMS aimed to explore the potential effects of TXA on intracranial bleeding and infarction. Specifically, it was hypothesised that TXA could reduce intracranial bleeding and/or increase cerebral infarction. Methods: The IBMS was nested within the CRASH-3 trial: a prospective, double-blind, parallel-arm, randomised trial. Patients eligible for the CRASH-3 trial, with a Glasgow Coma Scale (GCS) score of ≤ 12 or intracranial bleeding on pre-randomisation CT were eligible. Outcomes were examined on routinely collected brain scans done pre- and/or post-randomisation. The primary outcome is the volume of intra-parenchymal bleeding in patients randomised within three hours of injury. Secondary outcomes include new and progressive bleeding, post-neurosurgical bleeding, infarction, and a composite “poor outcome”. The primary outcome was analysed using a linear mixed model, and dichotomous outcomes using relative risks or hazard ratios. Findings: The IBMS included 14% of the CRASH-3 trial patients (n=1767/12,737): 884 TXA, 883 placebo. Patients had a median baseline GCS of 7 (IQR 3–10). Only 46% of patients were scanned pre- and post-randomisation (n=812/1767) and 35% were scanned post- but not pre-randomisation (n=614/1767). A total of 21% of patients had evidence of neurosurgical haemorrhage evacuation on a post-randomisation scan. There was no evidence for a reduction in intra-parenchymal bleeding with TXA (1.09, 95% CI 0.81–1.45) or in intracranial bleeding in neurosurgical patients (0.79, 95% CI 0.57–1.11). There was no evidence for a reduction in the composite (RR=1.01, 95% CI 0.93–1.10) or increase in the hazard of infarction with TXA (HR=1.31, 95% CI 0.95–1.80). In patients scanned pre- and post-randomisation, there was no evidence that TXA reduces progressive bleeding (RR=0.92, 95% CI 0.74–1.13) and no clear evidence that TXA reduces new bleeding (RR=0.86, 95% CI 0.72–1.02). Conclusions: Routine imaging cannot provide reliable information on the effects of TXA in TBI. The associated methodological flaws mean that the treatment effect estimates are not valid and precise. 1) The large proportion of missing post-randomisation scans could depend on whether a patient received TXA. 2) The inclusion of a large proportion of severely injured patients may dilute effect estimates towards the null. 3) The receipt of TXA may affect whether patients undergo neurosurgery, and this complicates the assessment of the effects of TXA using scans done post-randomisation and post-neurosurgery. Implications for future research: If a research protocol mandated that scans were done at a set time-point post-randomisation, this would reduce the risk of bias from missing outcomes. If less severely injured patients were included, this would reduce the occurrence of neurosurgery and missing outcomes as a result of death.